• An attention-driven image aesthetic assessment model has been proposed in a deep network consisting of three modules: spatial, channel and color. • A reinforcement learning framework for prompt improvement for increased aesthetic scores of images generated by Stable Diffusion. • Large language models for prompt editing to improve the aesthetics of generated images, guiding the actions in a reinforcement learning process. Generative models, e.g., stable diffusion, excel at producing compelling images but remain highly dependent on crafted prompts. Refining prompts for specific objectives, especially aesthetic quality, is time-consuming and inconsistent. We propose a novel approach that leverages LLMs to enhance prompt refinement process for stable diffusion. First, we propose a model to predict aesthetic image quality, examining various aesthetic elements in spatial, channel, and color domains. Reinforcement learning is employed to refine the prompt, starting from a rudimentary version and iteratively improving them with LLM’s assistance. This iterative process is guided by a policy network updating prompts based on interactions with the generated images, with a reward function measuring aesthetic improvement and adherence to the prompt. Our experimental results demonstrate that this method significantly boosts the visual quality of generated images when using these refined prompts. Beyond image synthesis, this approach provides a broader framework for improving prompts across diverse applications with the support of LLMs.

Driven by accelerated product obsolescence and frequent consumer replacements, electronic waste is growing rapidly. Waste recycling, as a core component of resource reuse, has become an important means of alleviating resource scarcity and reducing environmental pollution. In the process of recycling discarded products, the efficiency of disassembly operations is crucial. To improve disassembly efficiency and maximize resource utilization, this work proposes a hybrid disassembly line structure that incorporates both linear and U-shaped workstations. Shared labor is introduced between adjacent disassembly lines, allowing workers to flexibly execute tasks across lines. This resource-sharing mechanism enhances task coordination and reduces idle time, contributing to improved system efficiency. Using a precedence relationship graph to model dependencies among tasks, we develop a mathematical model aimed at maximizing profit. We use an exact solver to verify the model and adopt a variant of dueling deep Q-network, called PER-Dueling DQN (PDDQN), which incorporates prioritized experience replay to enhance sampling efficiency and solve the model optimally. A simulation environment aligned with this problem is constructed for the reinforcement learning agent. We compare the proposed method with other reinforcement learning approaches, including advantage actor-critic, proximal policy optimization, and trust region policy optimization. Through experiments on disassembling products of different sizes, the feasibility and effectiveness of PDDQN are demonstrated, exhibiting significant advantages over other methods.

Industry 5.0 promotes the transformation of manufacturing toward flexibility, personalization, and sustainability. As a critical component of closed-loop manufacturing systems, disassembly operations urgently require more flexible and efficient human–robot collaboration models. To this end, this work, for the first time, proposes a multihuman–robot collaborative circular disassembly line balancing problem. By allowing workers to move between robotic workstations, the proposed system enhances operational flexibility. Furthermore, a multiworker mechanism is introduced to improve fault tolerance and system stability, overcoming the limitations of fixed worker positions in existing collaborative disassembly research. To solve this problem, we formulate a discrete-time mixed-integer programming model based on product AND/OR graphs, aiming to maximize disassembly profit. The model’s correctness is verified using CPLEX. Additionally, we develop a heterogeneous graph neural network-enhanced proximal policy optimization (PPO) algorithm. By integrating product and workstation information into a heterogeneous graph, the algorithm performs two-stage feature extraction and node embedding via graph neural networks. Based on these embeddings, the agent dynamically selects multiple actions per decision step to simulate the behavior of multiple workers moving simultaneously. Experimental results show that the proposed method outperforms traditional reinforcement learning algorithms such as PPO and deep Q-network algorithm in terms of disassembly profit. Moreover, it demonstrates strong generalization capability in cross-task transfer and scalability experiments involving different task graph sizes. The improved performance is achieved with acceptable computational time.

As the demand for disassembling end-of-life products grows, limitations in traditional disassembly line design, low efficiency, and high resource consumption become increasingly evident. Particularly in large-scale disassembly tasks, where the cost of conventional remanufacturing rises and the technologies fail to meet high-efficiency requirements. The integration of robots into disassembly lines is a promising solution to alleviate these issues. This work presents a multi-product hybrid disassembly line balancing problem that considers machine wear rates and establishes a mixed-integer programming model guided by profit maximization to address it. An improved fireworks algorithm is used in the proposed approach. The developed solution is compared with genetic and ant colony algorithms. Evaluation results and analysis demonstrated the competitive efficiency and stability of our approach.

Generative models such as stable diffusion have recently achieved significant success in producing high-quality images conditioned on textual inputs. It is difficult to control the quality of generated images, such as aesthetic quality, after an image generative model has been trained. In this paper, we pro-pose a novel method that leverages reinforcement learning (RL) to enhance the aesthetic appeal of images produced by stable diffusion models. An aesthetic assessment model has been first developed by modelling influence factors of image aesthetics in a deep network and then trained on publicly available datasets. The assessment model produces an aesthetic score of an image serving as the reward function in the RL framework. By reframing the denoising process of the stable diffusion model as a sequential decision-making problem, the intermediate denoising steps can be formulated as actions in a Markov decision process (MDP). The proximal policy optimization (PPO) algorithm can be applied in the RL framework to optimize the network parameters (i.e., U-Net) in the stable diffusion model, aiming to maximize the expected aesthetic reward. Through extensive experiments, we demonstrate that the proposed method can significantly improve the aesthetic quality of the generated images while maintaining their diversity and adherence to input prompts.

The disassembly and assembly line balancing problem (DALP) is a critical task in industrial production, involving the efficient organization of disassembly and assembly tasks to improve the productivity and flexibility of production lines. In practical applications, task allocation, robot movement, and workstation layout optimization are key factors affecting production efficiency. This study proposes an improved parallel advantage actor-critic algorithm to address DALP with space constraints due to robot movement. Considering the limitations of workstation space, this approach optimizes the robot's movement paths between workstations, reducing the cost of opening workstations, and optimizing task allocation strategies. To enhance the convergence speed and stability of the conventional Parallel A2C algorithm, action space optimization and a greedy strategy are incorporated into the algorithm. Experimental results demonstrate that the improved parallel advantage actor-critic outperforms the A2C and AC algorithms in terms of efficiency and performance, particularly in handling disassembly tasks with space constraints, significantly improving the operational efficiency and economic benefits of the production line.

Product disassembly is significant for recycling scrapped products and reducing environmental pollution and resource waste. The recovery, reuse, and recycling of industrial products is crucial in modern industry. Manual disassembly efficiency significantly impacts the disassembly line’s overall effectiveness, especially workers’ skill level and learning efficiency. This paper proposes a multi-period personnel scheduling problem that considers worker learning effects. A mixed integer programming model for the disassembly balance problem was established to maximize disassembly profit. This problem is solved using a new reinforcement learning algorithm, the importance-weighted actor-learner architecture (IMPALA). The correctness and effectiveness of the proposed algorithm are verified through comparative experiments with the famous IBM optimizer CPLEX and some popular peer algorithms.

The advancement of human–robot collaboration technology has positioned remanufacturing as a crucial part of the circular economy, driving both economic growth and environmental sustainability. In the era of Industry 5.0, these technologies enhance the efficiency and flexibility of disassembly tasks. However, most research on human–robot collaborative disassembly (HRCD) line balancing overlooks the mobility of workers. This study introduces a profit-oriented HRCD model incorporating mobile workers. To address large-scale HRCD challenges, it proposes a dynamic attraction rate mechanism that improves the traditional carnivorous plant algorithm (CPA), tackling issues of slow convergence and local optimization. The experimental framework includes three validation phases: 1) comparison with the exact solver IBM ILOG CPLEX Optimization Studio (CPLEX); 2) parameter sensitivity analysis; and 3) benchmarking against seven state-of-the-art algorithms. Results demonstrate that HRCD with mobile workers significantly boosts disassembly efficiency and reduces disassembly time compared to traditional methods. Additionally, it increases profits through flexible task allocation. In cases of incomplete disassembly, HRCD with mobile workers yields an average benefit increase of 87.64% over conventional disassembly modes. A comparative evaluation with other swarm intelligence algorithms further highlights the superior solution quality and time efficiency of the improved CPA.

Many mobile applications have resorted to deep neural networks (DNNs) because of their strong inference capabilities. Since both input data and DNN architectures could be sensitive, there is an increasing demand for secure DNN execution on mobile devices. Towards this end, hardware-based trusted execution environments on mobile devices (mobile TEEs), such as ARM TrustZone, have recently been exploited to execute CNN securely. However, running entire DNNs on mobile TEEs is challenging as TEEs have stringent resource and performance constraints. In this work, we develop a novel mobile TEE-based security framework that can efficiently execute the entire DNN in a resource-constrained mobile TEE with minimal inference time overhead. Specifically, we propose a progressive pruning to gradually identify and remove the redundant neurons from a DNN while maintaining a high inference accuracy. Next, we develop a memory optimization method to deallocate the memory storage of the pruned neurons utilizing the low-level programming technique. Finally, we devise a novel adaptive partitioning method that divides the pruned model into multiple partitions according to the available memory in the mobile TEE and loads the partitions into the mobile TEE separately with a minimal loading time overhead. Our experiments with various DNNs and open-source datasets demonstrate that we can achieve 2-30 times less inference time with comparable accuracy compared to existing approaches securing entire DNNs with mobile TEE.

For decades, one-time verification has been the standard for user verification at entry points, office rooms, etc. However, such approaches request users to provide their secrets (e.g., entering passwords and collecting fingerprints) and re-verify (e.g., screen shutdown) manually. Thus, they cannot confirm whether the user is a legitimate or an imposter after verification, which raises the urgent demand for a more convenient and secure solution to perform continuous user verification. However, existing continuous verification methods heavily rely on users' active participation, which is inconvenient. Toward this end, we propose a continuous user verification system, BioTag, which utilizes the low-cost radio frequency identification (RFID) technology to capture unique physiological characteristics rooted in the users' respiration motions for continuous user verification. Specifically, we use two RFID tags attached to a user's chest and abdomen to capture the user's intrinsic respiratory patterns via RFID signals. We develop respiratory feature extraction methods based on waveform morphology analysis and fuzzy wavelet transformation (FWPT) to derive unique biometric information from the user's respiration signals. Furthermore, we develop an adaptive classifier using the gradient boosting decision tree (GBDT) to identify legitimate users and attackers accurately. Extensive experiments involving 41 participants demonstrate that BioTag can robustly authenticate users and detect various types of adversaries with low training effort. In particular, our system can achieve over 95.2% and 94.8% verification accuracy on random attack and imitation attack scenarios, respectively.

Deep learning (DL) has demonstrated great performance in various applications on powerful computers and servers. Recently, with the advancement of more powerful mobile devices (e.g., smartphones and touch pads), researchers are seeking DL solutions that could be deployed on mobile devices. Compared to traditional DL solutions using cloud servers, deploying DL on mobile devices have unique advantages in data privacy, communication overhead, and system cost. This article provides a comprehensive survey for the current studies of adopting and deploying DL on mobile devices. Specifically, we summarize and compare the state-of-the-art DL techniques on mobile devices in various application domains involving vision, speech/speaker recognition, human activity recognition, transportation mode detection, and security. We generalize an optimization pipeline for bringing DL to mobile devices, including model-oriented optimization mechanisms (e.g., pruning and quantization) and nonmodel-oriented optimization mechanisms (e.g., software accelerator and hardware design). Moreover, we summarize popular DL libraries regarding their support to state-of-the-art models (software) and processors (hardware). Based on our summarization, we further provide insights into potential research opportunities for developing DL for mobile devices.

Neuron networks pruning is effective in compressing pre-trained CNNs for their deployment on low-end edge devices. However, few works have focused on reducing the computational cost of pruning and inference. We find that existing pruning methods usually remove parameters without fine-grained impact analysis, making it hard to achieve an optimal solution. This work develops a novel mixture pruning mechanism, MIXP, which can effectively reduce the computational cost of CNNs while maintaining a high weight compression ratio and model accuracy. We propose to remove neuron bond that can effectively reduce convolution computations and weight size in CNNs. We also design an influence factor to analyze the importance of neuron bonds and weights in a fine-grained way so that MIXP could achieve precise pruning with few retraining iterations. Experiments with MNIST, CIFAR-10, and ImageNet datasets demonstrate that MIXP could achieve significantly fewer FLOPs and retraining iterations on four widely-used CNNs than existing pruning methods.